Import Error for 'MetaPath2Vec'

Questions
#1

How can I fix this import error:
ImportError: cannot import name ‘MetaPath2Vec’ from ‘dgl.nn.pytorch’ (C:\Users\USER\anaconda3\lib\site-packages\dgl\nn\pytorch_init_.py)

I updated the dgl by this command:
conda install -c dglteam dgl

my pytorch version: 1.13.0
my dgl version: 0.9.1

#2

Hi @zara , DGL 0.9.1 does not have MetaPath2Vec, which will be available in DGL 1.0. A workaround to use it for now is to build the DGL master branch from source.

#3

Thank you for your reply. I used source codes that were here:
https://docs.dgl.ai/en/latest/_modules/dgl/nn/pytorch/network_emb.html#MetaPath2Vec
but I have this Error:
RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu! (when checking argument for argument mat1 in method wrapper_addmm)

Actually I used cuda in my implementations and after I used metapath2vec source codes I faced that Error.
I changed the cpu device of this sorce code to cuda but it doesn’t work.

#4

Did you run it on DGL 0.9.1?

#5

yes, I ran it on DGL 0.9.1 and I faced that RuntimeError.

#6

RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu! (when checking argument for argument mat1 in method wrapper_addmm)

Actually I used cuda in my implementations and after I used metapath2vec source codes I faced that Error.
I changed the cpu device of this sorce code to cuda but it doesn’t work.

Are your model and graph on the same device?

#8

when I want to put my graph on cuda, I faced this Error:
C:\Users\Administrator\dgl-0.5\src\runtime\c_runtime_api.cc:88: Check failed: allow_missing: Device API gpu is not enabled. Please install the cuda version of dgl.

I checked and cuda was available.

#9

Please check this page to install DGL with corresponding CUDA version. You might need to manually uninstall the old version first.

#10

thank you mufeili for your response.
unfortunately my cuda version is 11.7 and the link you sent doesn’t contain installing dgl with cuda version 11.7
Should I uninstall the cuda version 11.7 and install cuda version 11.6?
Can you help me and say how to do that?

#11

You can install cuda 11.6 by installing the corresponding PyTorch version if needed.

#12

I did it and I have another problem and i face cuda out of memory.

GPU 0 : Intel(R) UHD Graapics - (GPU Memory: 7.9 and Shared GPU memory: 7.9)

GPU 1 : NVIDIA GeForce RTX 3050 Ti Laptop GPU - (Dedicated GPU memory: 4.0 and GPU Memory: 11.9 and Shared GPU memory: 7.9)

I ran a code and I faced this error:

DGLError: [00:55:04] C:\Users\Administrator\dgl-0.5\src\runtime\cuda\cuda_device_api.cc:119: Check failed: e == cudaSuccess || e == cudaErrorCudartUnloading: CUDA: out of memory

I searched and find out my codes run on cuda : 0 (I didn’t tell in codes)

I search and see: GPU CUDA compute capability major version: 0

I think my codes run on GPU 0

How can I run my codes on GPU 1?

How to install cuda on GPU 1?

#13

CUDA_VISIBLE_DEVICES=1 python xxx.py

#14

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.